[LoRA模型] 绪儿-【解决手崩难题】整合包
模型简介：这个LORA针对SD画不好手做了强化训练，绪儿已成精的个人空间_哔哩哔哩_bilibili
我测试了大部分热门模型，每100张图坏手的概率小于15%。
基本不会坏手。同时整合了一个新的脸型。一款中式服装。
希望对修手烦恼的人有帮助。LORA权重推荐0.7-1.
需要注意的是这个LORA，脸的精度太高。
画半身肖像的时候不要开ADE修脸插件。会导致脸变形。
如果想脸更高清，可以在跑完图以后开ADE局部重绘脸。
画全身肖像可以开ADE没问题。
测试提示词和参数：
prompt:
(dramatic, gritty, intense:1.4),masterpiece, best quality, 32K，insane details, intricate details, hyperdetailed, hyper quality, high detail, ultra detailed, Masterpiece, (full body:1.4),Red eyes，(white hair:1.6)，(braid:1.3)，,an girl in a red dress is in front of white eyes, Unreal background, smoke and mist, like something coming out. Chinese Taoist ghost legend，horror-inspired, Chinese Horror，dollcore, dark violet and light blue, contrast between light and dark，photo-realistic hyperbole, japonism influenced pieces, cryptidcore，Junji Ito, miho hirano, ilya kuvshinov, Yimou Zhang, Thick paintings, oil paintings, ink drawings，cinematic textures, top view, extreme close-ups, REALISTIC 3D, ULTRA DETAILED. Blender, C4D, high detail, high resolution,,A shot with tension，(sky glows red,Visual impact,giving the poster a dynamic and visually striking appearance:1.2),Chinese Zen style,impactful picture,,<lora:绪儿-红衣女鬼:0.8>
Negative prompt:
bad-artist， bad-artist-anime， bad-hands-5， bad-image-v2-39000， bad-picture-chill-75v， bad_prompt， bad_prompt_version2， badhandv4， NG_DeepNegative_V1_75T， EasyNegative，2girls, 3girls,,(nsfw:1.3),
Model:
梦境夢の世界dreamWorld_v15novae
参数：
Steps: 55, Size: 1624x2496, Seed: 3952079349, Sampler: DPM++ 2M SDE Karras, CFG scale: 7
页:
[1]